Microsoft will end Windows 10 support on October 14, 2025. After this date, Microsoft 365 apps will no longer be compatible with Windows 10, and users must upgrade to Windows 11 for continued Office functionality and security updates. Despite low Windows 11 adoption, Microsoft offers extended security updates for Windows 10 users at a cost. TPM 2.0 is required for Windows 11, posing challenges for older hardware.

London-based Synthesia raises $180M at a $2.1B valuation for its AI-driven video generation platform. The Series D round was led by NEA, doubling its valuation since 2023. Synthesia enables creating realistic human avatars for videos in over 30 languages and plans to introduce interactive features and more expressive avatars.

CohereAI's Aya Expanse is a multilingual model series designed to excel in 23 languages. The 32B parameter version outperforms top models like Gemma 2 27B and Llama 3.1 70B in multilingual tasks. It offers advanced capabilities and is available for testing via HuggingFace, demonstrating strong performance without needing retraining.

Vidu 2.0, an advanced AI video generation platform, enhances efficiency with a 4-second, 512P clip in under 10 seconds at 4 cents per second. It offers improved visual consistency, smoother transitions, and off-peak unlimited generation. The platform now supports simultaneous creation of up to four clips and has launched a Chinese site with localized UI and payment methods for better user experience.

iFLYTEK launches Xinghuo Deep Reasoning Model X1, the first large model on a domestic computing platform featuring advanced reasoning capabilities. It excels in Chinese mathematics and has been successfully deployed in education and healthcare, achieving top-tier results in various tests. An upgraded medical model is planned for release this year.

Samsung's Galaxy Unpacked 2025 on Jan 22 in San Jose will unveil the Galaxy S25 series, including S25, S25+, S25 Ultra, and possibly S25 Slim, powered by Snapdragon 8 Elite. AI advancements and new Galaxy Ring 2 and smart glasses expected. One UI 7 to debut.

OpenAI's o1 model reasons in multiple languages including Chinese, sparking industry discussions. Experts speculate on causes ranging from multilingual training datasets to third-party data annotation services, and the model possibly choosing the most effective language to handle "hallucination" issues during reasoning.

Google's Gemini AI system demonstrates advanced capability in processing multiple visual streams simultaneously through an experimental app called AnyChat. Unlike other AI platforms such as ChatGPT, Gemini can handle both real-time videos and static images without compromising performance. This breakthrough, achieved by optimizing Gemini's attention mechanism, has broad application prospects in healthcare, engineering, quality control, education, and creative fields. While currently available on

Nvidia invests in Taiwan's MetAI, a startup creating AI-driven digital twins for industrial applications. MetAI's tech swiftly transforms CAD files into 3D environments, enhancing robot training and validation. The company raised $4M in seed funding, aiming to expand R&D and accelerate market entry. Nvidia sees potential in MetAI's scalable AI training environments for advanced automation.

The US President signed an administrative directive to accelerate AI infrastructure development, aiming to enhance national security and economic competitiveness. The directive pushes for faster data center construction and clean energy initiatives. Meanwhile, the US and India agreed on mutual AI investments. Former President Trump criticized the move, advocating for more innovation-friendly policies.

In an AI copyright case, Meta's internal communications reveal their commitment to outperforming competitors like OpenAI and Anthropic with Llama 3. Despite using suboptimal datasets, including copyrighted material, Meta aims to lead the industry. Zuckerberg stated Llama 3 is competitive and expects future models to be the most advanced. Mistral is viewed dismissively. Data acquisition pressures are intense.

Apple joins Ultra Accelerator Link Consortium to optimize AI chip connections in data centers. UALink standard supports GPUs and custom chips enhancing AI training and execution. Apple develops new server chips for AI efficiency. Nvidia absent due to NVLink technology. Apple updates AI news summary feature after inaccuracies.

Nvidia's CEO believes truly useful quantum computers are decades away, but Nvidia invests in quantum computing through partnerships like SEEQC. SEEQC recently raised $30 million for developing efficient quantum processors and managing qubits, crucial for enterprise needs. They aim to reduce quantum hardware complexity and improve integration with data centers. This aligns with growing demand for resilient chip supply chains and involves collaborations with companies like BASF and Merck to accele

OpenAI's ChatGPT introduces task scheduling and automated reminders, currently in beta for Plus, Team, and Pro users. Features include one-time or recurring alerts via chat commands, managed on a dedicated "Tasks" page, with up to 10 active tasks supported during the trial phase. Notifications are sent through push or email, aiming to enhance user productivity and system autonomy. Compatibility with Windows expected by Q1.

MiniMax releases MiniMax-01 series with 4M token context window, outperforming competitors like Gemini 1.5 Pro. Available on Hugging Face and GitHub, these models offer cost-effective API access and innovative Lightning Attention for efficient long-context processing. Open-sourcing aims to advance AI agent capabilities.

Salesforce CEO Marc Benioff criticizes Microsoft's AI efforts, particularly Copilot, as repackaged OpenAI tech lacking significant improvements. He claims customers rarely use Copilot and compares it to an impractical Clippy 2.0. Benioff asserts Salesforce leads in AI transactions and Agentforce sets industry standards. Internal reports suggest Microsoft struggles with Copilot's functionality, security issues, and relies on third-party vendors, causing data breaches and operational gaps. Despite

Microsoft's AI Red Team released a white paper on security challenges of generative AI systems, proposing strategies to address these risks. Key findings include: generative AI introduces new attack vectors like prompt injection; human expertise remains crucial for comprehensive risk assessment; and a defense-in-depth approach combining continuous testing and robust defenses is essential for mitigating risks. Traditional security issues persist in AI solutions, emphasizing the need for integrate

Advancements in AI reasoning models like OpenAI's o1 have spurred new innovations and competitors such as DeepSeek's R1 and Google's Gemini 2. These models use chain-of-thought techniques for better accuracy but come at a higher cost, sparking debate on their value. Users are advised to provide detailed briefs rather than simple prompts to unlock the models' full potential, allowing for autonomous problem-solving and improved outcomes. For non-reasoning LLMs, refining prompt methods can also enh

Microsoft forms CoreAI - Platforms and Tools Division led by Jay Parikh to integrate AI across its Developer Division and AI Platform team. Aim to create an AI-first application stack focusing on Azure as foundational infrastructure for developing intelligent agents and AI applications for internal and external use. This move marks a significant step in Microsoft's strategy to reshape application categories through advanced AI platforms and tools.

MBZUAI's LlamaV-o1 model sets new standards in multimodal AI reasoning, emphasizing step-by-step inference over final answers. It excels in tasks requiring precision and transparency, such as medical imaging and financial chart interpretation. Trained on LLAVA-CoT-100k, it scored 68.93 on VRC-Bench, outperforming competitors. The model's focus on explainability enhances trust in critical sectors like healthcare and finance. However, its performance is limited by data quality and should not be us